Parallel stochastic search for protein secondary structure prediction

Abstract

Prediction of the secondary structure of a protein from its aminoacid sequence remains an important and difficult task. Up to this moment, three generations of Protein Secondary Structure Algorithms have been defined: The first generation is based on statistical information over single aminoacids, the second generation is based on windows of aminoacids-typically 11-21 aminoacids- and the third generation is based on the usage of evolutionary information. In this paper we propose the usage of naïve Bayes and Interval Estimation Naïve Bayes (IENB) -a new semi naïve Bayes approach- as suitable third generation methods for Protein Secondary Structure Prediction (PSSP). One of the main stages of IENB is based on a heuristic optimization, carried out by estimation of distribution algorithms (EDAs). EDAs are non-deterministic, stochastic and heuristic search strategies that belong to the evolutionary computation approaches. These algorithms under complex problems, like Protein Secondary Structure Prediction, require intensive calculation. This paper also introduces a parallel variant of IENB called PIENB (Parallel Interval Estimation Naïve Bayes).